Overview contents
Research question
In which environments and conditions can a visual map provide sufficiently reliable positioning to support navigation, augmented reality or autonomous systems?
Technical context
During mapping, images of the environment are captured and relatively stable visual features are identified and linked to known coordinates. During use, the system searches the map for areas that match what the camera sees.
The expected result is an estimate of the device’s pose: three-dimensional position and orientation, or six degrees of freedom. Inertial sensors and SLAM systems can maintain tracking between visual localisation updates.
Google Geospatial is an example of a global-scale VPS based on Street View. For private indoor environments, we instead study purpose-built maps and dedicated visual localisation methods.
What we are studying
- 01
Capture and quality control of visual maps for private environments.
- 02
Relocalisation, pose estimation and continuity of tracking.
- 03
Fusion with GPS, Wi-Fi, BLE and inertial sensors to narrow down the candidate area.
- 04
Map updates and image processing in sensitive settings.
Applications to explore
- Indoor navigation at trade fairs, hospitals, campuses, airports and car parks.
- Augmented reality content and instructions anchored to physical points.
- Contextual recognition of machinery and maintenance procedures.
- Experimental relocalisation of robots and drones in areas without GPS coverage.
How it can combine with other technologies
Wi-Fi, BLE and inertial sensors can provide an initial position or maintain an estimate when the camera cannot see recognisable elements. VPS can refine the pose. A 3D reconstruction can provide the environment’s visual context, but does not automatically replace the visual-feature map required for localisation.
What a prototype needs to establish
- V01
Success rate and time needed to obtain the first valid localisation.
- V02
Position and orientation error in the chosen real setting.
- V03
Continuity after occlusion, crowds, lighting changes or loss of tracking.
- V04
The cost and frequency of map updates.
- V05
Latency, energy consumption, privacy and differences between devices.
Known limitations and risks
Repetitive corridors, featureless surfaces, occlusion and changes to furnishings can reduce reliability.
Every site requires initial capture and a maintenance process; there is no single accuracy level valid for every environment.
Technical references